Manufacturing Revolutions: Industrial Policy and Industrialization in South Korea*
Nathan Lane†
Current Version: September 2021 [First Draft: October 2016]
Abstract
I study the impact of industrial policy on industrial development by considering a canonical intervention. Following a political crisis, South Korea dramatically altered its development strategy with a sector-specific industrial policy: the Heavy and Chemical Industry (HCI) drive, 1973-1979. With newly assembled data, I use the sharp introduction and withdrawal of industrial policies to study the impacts of industrial policy—during and after the intervention period. I show (1) HCI promoted the expansion and dynamic comparative advantage of directly targeted industries. (2) Using variation in exposure to policies through the input-output network, I show HCI indirectly benefited downstream users of targeted intermediates. (3) I find direct and indirect benefits of HCI persisted even after the end of HCI, following the 1979 assassination of the president. These effects include the eventual development of directly targeted exporters and their downstream counterparts. Together, my findings suggest that the temporary drive shifted Korean manufacturing into more advanced markets and created durable industrial change. These findings clarify lessons drawn from South Korea and the East Asian growth miracle. JEL: L5 O14 O25 N6. Keywords: industrial policy, East Asian miracle, economic history, industrial development, Heavy-Chemical Industry Drive, Heavy and Chemical Industry Drive.
1 Introduction
Miracles by nature are mysterious. The forces behind the East Asian growth miracle are no exception. Industrial policy (IP) has defined Asia’s striking postwar transformation (Rodrik 1995). The ambitious
*I benefited from conversations with Daron Acemoglu, Robert Allen, Sam Bazzi, Sascha O. Becker, Nils Bohr, Timo Bop- part, David Cole, Arin Dube, Samantha Eyler-Driscoll, Alice Evans, Mounir Karadja, Max Kasy, Oliver Kim, Changkeun Lee, Weijia Li, Ernest Liu, Andreas Madestam, Javier Mejia, Matti Mitrunen, Aldo Musacchio, Suresh Naidu, Nathan Nunn, Veronica Perez, Dwight Perkins, Pseudorasmus, Erik Prawitz, Pablo Querubin, Dani Rodrik, Martin Rotemberg, Todd Tucker, Eric Verhoogen, Robert Wade, and Lisa Xu. I would like to thank audiences at American University, Australia Na- tional University Economics, Australia National University—Crawford School, CSAE Oxford, College de France, Geneva Graduate Institute Blended Finance Conference, EBRD, European Econometric Society Summer and Winter Meetings, Har- vard, IMT Lucca, Institute of New Structural Economics–Peking University, INSEAD, Kellogg School of Management, Ko- rean Development Institute, LSE ID, MIT, NBER SI, Nottingham University, NYU-Abu Dhabi, Univ. of Oxford, OzClio, Seoul National University, Sussex University, UMASS–Amherst, University of Melbourne, UNSW, University of Technology Sydney, and University of Wollongong for their helpful comments. I would especially like to thank my committee: Melissa Dell, Torsten Persson, James Robinson, and David Stromberg. This study was made possible with excellent assistance from Kim Chan, BoSuk Hong, Veronic Perez, Cheong Yeon Won, and Hye Jin Won. I would also like to thank the staff of the Bank of Korea and the Korean Development Institute. †University of Oxford and SoDa Labs (Monash). See nathanlane.info or contact me at [email protected]
1 development strategies pursued across this region have shaped global industrial strategy, from Southeast Asia to Sub-Saharan Africa (Rodrik 2005; Robinson 2010; Lin 2012). With rare exception, every developing country has pursued some type of IP intervention. Early development economists argued that these strategies play a fundamental role in industrialization (Rosenstein-Rodan 1943; Hirschman 1958). Others argued IP is inherently deleterious (Baldwin 1969; Krueger 1990), and its role in East Asia was counterproductive (Pack 2000). While industrial policies have re-entered the broader policy arena, empirical evidence surrounding their failures and successes in the developing world is scant.1 This holds for an episode associated with their use, the East Asian miracle.
South Korea embodied the transformation of East Asian economies. The economy entered the 1960s as an un- stable industrial laggard. By the 1980s it had undergone a manufacturing transformation that had taken West- ern nations over a century to achieve (Nelson and Pack 1998).2 How did South Korea transition from a light export economy to an industrial powerhouse?
I explore the role of industrial policy in Korea’s transformation. By industrial policy, or IP, I mean something specific: deliberate state action intended to shift the composition of national economic activity (Lindbeck 1981; Chang 2003; Noland and Pack 2003). In developing countries, like South Korea in the 1980s, this shift is hoped to be growth enhancing. To study IP I turn to a key South Korean intervention that sought to change the in- dustrial trajectory of the small, open economy.
The policy in question is South Korea’s Heavy and Chemical Industry (HCI) drive, 1973 to 1979. HCI was a definitive postwar industrial strategy. In some dimensions, such as its emphasis on industry spillovers and capital accumulation, HCI shared similarities with early postwar policy-making (Rosenstein-Rodan 1943; Nurkse 1953). In other dimensions, it was synonymous with the East Asian miracle (Vogel 1991). Its outward orientation resembled policies pursued by contemporaries. HCI was itself inspired by Japan’s earlier policy, and resembled contemporaries, like Taiwan (Cheng 1990, 2001). Korea’s own industrial drive influenced strategies across the globe, as economies, such as Malaysia, “looked east” for developmental templates. The varied record of imitators, however, has fueled HCI’s notoriety.3 Among industrial policies, HCI looms large.
This study aims hopes to address some empirical issues surrounding industrial policy. Like many controver- sial East Asian policies—and infant industry policy broadly—evidence around its efficacy is incomplete (Lane 2020). This study confronts two challenges that complicate analysis of infant industry IP, in particular research design. Using newly assembled data, I also confront empirical obstacles to studying the East Asian miracle.
For over a century economists have discussed the empirical obstacles to studying industrial policy (Meredith 1906; Grubel 1966). Theoretically, most optimal policies are temporary, and justifications rely on assisting sec- tors with either dynamic comparative advantage (Greenwald and Stiglitz 2006) and (or) spillovers (for exam- ple, inter-industry linkages or across firm learning) (Hirschman 1958; Grossman 1990). Tests of theoretical jus- tifications, however, are moot against many unobserved political realities surrounding IP (Rodrik 2005, 2012). These political confounders mean that IP often goes to sub-optimal and politically motivated recipients. These same political realities also mean that IP is seldom temporary (Juhasz 2018). Last, political confounders mean that de jure sectoral policy may not translate into de facto policy. For these reasons, I argue that the HCI context provides an important vehicle for studying the impact of industrial policy on industrial development.
I use variation introduced by South Korea’s HCI episode to estimate the impact of purposeful IP on short-run, as well as persistent, Korean industrial development. I do so by using variation introduced by the HCI context.
1Recent contributions exceptions are discussed below. See Lane (2020) for review of the current empirical literature and policies. 2According to the Penn World Tables, in 1960 South Korea’s per capita national income lagged behind Cameroon, the Central African Republic, Haiti, Madagascar, Morocco, Niger, and Tanzania (Werlin 1991; Feenstra, Inklaar, and Timmer 2015). 3For global experience of HCI-style policies, see Kim et al. (2013); Moreira (1994); Lall (1995); Lall (1996).
2 I argue that external politics precipitated HCI’s start in 1973 and its termination in 1979. President Nixon’s promise to withdraw U.S. forces from the Asia-Pacific area shook regional allies. Like Southern Vietnam, the Republic of Korea relied on U.S. support against Communist-backed North. Washington’s about-face catalyzed South Korea to incubate a heavy industrial complex. Rigorously implemented under the duress of crisis, the drive targeted strategic, yet feasible, infant industries (Stern et al. 1995; B.-k. Kim and Vogel 2011). However, this drive was temporary. The assassination of the president in 1979 ended his regime’s core project. I argue this variation is a useful natural experiment for estimating the impacts of IP on Korea’s industrial development.
Studying the HCI experience entails constructing new data on industrial outcomes spanning South Korea’s miracle period (1967-1986). I do so by harmonizing material from archival sources, digitized industrial sur- veys, and vintage machine-readable statistics into consistent panel data. Importantly, I combine industry-level data with digitized input-output accounts. The result is a data set spanning a key episode of East Asian devel- opment.
HCI’s setting provides an estimation strategy. I study the impact of IP by comparing changes in outcomes between targeted versus non-targeted manufacturing industries each year, before and after, the HCI announcement. This dynamic differences-in-differences (DID) strategy captures the impacts of HCI policies (investment incentives and trade policy). Pre-trends represent Korea’s counterfactual sectoral structure. Absent these interventions, industries would have evolved according to their pre-1973 pattern—or static comparative advantage. Post-1973 differences reveal the efficacy of IP in promoting sectors where South Korea had unrealized potential—or latent comparative advantage. I estimate these impacts using traditional two-way fixed effect (TWFE) estimators, as well new semiparametric estimation procedures (Callaway and Sant’Anna 2020; Sant’Anna and Zhao 2020).
My empirical strategy allows me to examine two principal justifications of industrial policy (See: Krueger and Tuncer 1982). First, by comparing the evolution of treated versus non-treated industries after Park’s assassina- tion, I confirm if that infant industry interventions were durable. In fact, I show some aspects of the drive only fully emerge after the policy drive. In doing so, I test for the dynamic impacts of IP. Second, another motiva- tion for IP is that benefits accrue to industries external to treated sectors. I explore this by estimating the im- pact of IP on industries differentially exposed to targeted sectors through industrial linkages. With measures constructed from historical input-output accounts, I compare the evolution of (non-targeted) industries with weak links to industries with strong linkages to HCI sectors.
I highlight three findings. First, I show significant, positive impacts of IP across industrial development out- comes in targeted (treated) industries. Relative to pre-HCI levels, HCI industries expanded their output over 100 percent more than non-treated sectors; labor productivity is over 60 percent higher. I show the relative ex- pansion of HCI industry does not result from a decline in non-treated industries. I find significant effects of IP on employment growth and export performance, and show output prices were 10 percent lower for HCI industry after 1973.
Second, the impact of HCI on industrial transformation is durable, and HCI promotes the long-run, dynamic comparative advantage of targeted export industry. Post-1979, industrial development outcomes—such as the share of economic activity—remained significantly higher compared to counterfactual industry. Moreover, HCI targeting is associated with the striking expansion of export industry. I estimate that HCI products were 13 percent more likely to achieve comparative advantage in global export markets after 1973. The revealed comparative advantage of HCI products increased 30 percent more than other manufacturing exports over the same period. Event study estimates show this ascent came to a head after the end of the policy. Hence, policies may have dynamic, long-run benefits beyond the planning period.
3 Third, HCI policies positively impacted the development of external, downstream industry, and promoted longer-run comparative advantage among downstream exporters. Downstream sectors with strong linkages to HCI industry expanded during the HCI period. During the drive, comparative advantage emerged among downstream exporters, yet the advantage fully materialized after 1979. Conversely, I show the backward- linkage effects of IP were, at best, limited. Thus, I find HCI supported development through supply-side effects passed through forward linkages. This comports with recent work on optimal interventions in networks by Liu (2019), who uses data from this paper to show that upstream sectors targeted by Korea correspond to sufficient statistics for industrial targeting.
Additionally, I show policies defy some characterizations of East Asian industrial policies and HCI. I do not find that HCI relied on strong, nominal trade barriers to protect output markets. Moreover, I do not find that HCI resembled import substitution (ISI) policies to which it has been compared. Rather, I find the HCI bundle likely operated through investment subsidies, promoting the capital formation and financing intermediates in targeted industry.
With these results I contribute to three literature. First, I build on nascent studies which use contemporary econometrics to study the efficacy of IP. This includes Nunn and Trefler (2010)’s work on structure of trade policy and industrial development, and recent country case-studies by Aghion et al. (2015) and Criscuolo et al. (2019). A complementary, parallel literature in empirical IO has begun articulating a structural framework for analyzing sector-specific policy bundles (see thoughtful work by Kalouptsidi (2018) and Barwick, Kalouptsidi, and Zahur (2019)).4 In development, work by Rotemberg (2017) and Martin, Nataraj, and Harrison (2017) have explored issues at the heart of IP through the experience of Indian SME policy. Though industrial policy is ubiquitous in practice, the literature remains notoriously underdeveloped (Lane 2020).
I contribute to the empirical study of industrial policy through natural experiments and historical case studies. I join Juhasz (2018), who uses Napoleonic Blockade to test for infant industry policy in historical France. Related work by Inwood and Keay (2013) and Harris, Keay, and Lewis (2015) study the developmental effects of output market protection with Canada’s early tariff experiments. Like this work, I find long-run effects of industrial policy, but in a contemporary setting and through contemporary export-oriented policies . These results dovetail with work studying dynamic comparative advantage and temporary natural experiments (Hanlon 2020; Mitrunen 2019; Pons-Benaiges 2017).5 Thus, I join historical work highlighting the potential of transitory policy to promote the evolution of “sunrise” industry. I do so by examining a purposeful, targeted intervention in a modern context. In analyzing intentionally targeted policies, I speak to a growing body of work evaluating place-based IP, which uses variation introduced by institutional aspects of policy. Notably, Criscuolo et al. (2019), use exogenous variation to study the impact of IP support targeted at lagging (UK) geographies; and Becker, Egger, and Ehrlich (2010) for lagging EU industry. I show the impact of policies aimed at sunrise industries in a (former) developing country setting, rather than lagging regions.
Second, I contribute to unresolved debates on the role of industrial policy and development, specifically con- troversies around East Asian miracle. Influential qualitative work emphasized the role of IP in newly industri- alizing economies.6 Wade (1990) and Amsden (1992) argue IP was vital to Taiwan’s and South Korea’s ascent. A sizable economics literature emphasized the implicit flaws of infant industry interventions (Baldwin 1969; Krueger 1990; Lal 1983).7 Economists challenged the lessons gleaned from East Asia, specifically with respect to targeting (Weinstein 1995; El-Agraa 1997; Lawrence and Weinstein 1999). Krueger (1995) and Pack (2000)
4Kalouptsidi (2018) shows the potential to use theoretically-grounded inference to detect commonly unobserved indus- trial policies. This IO literature shares likeness with an the earlier “new trade” literature, which used calibration exercises to study the impacts of infantry industry interventions (Baldwin and Krugman 1988; Head 1994; Irwin 2000). 5Hanlon (2020) studies the initial cost advantages of early steel shipbuilders, while Mitrunen (2019) examines the im- pact of Stalin’s export reparations policy on Finnish industry. For temporary government procurement policy, technology and managerial training, see Jaworski and Smyth (2018) and Giorcelli (2019). 6The qualitative literature is vast. See seminal work by Johnson (1982); Wade (1990); Vogel (1991); Amsden (1992); Evans (1995); Chibber (2002); and Kohli (2004) 7See: extensive critical discussions by Pack and Saggi (2006) and Noland and Pack (2003).
4 contend newly industrialized countries developed despite industrial strategy. Yoo (1990) argues HCI may have, in fact, harmed South Korea’s export performance relative to contemporaries.
Correlation studies of East Asia have shown a negative relationship between interventions and industry devel- opment, and argue that IP did not target high-spillover sectors (Lee 1996; Beason and Weinstein 1996; Noland 2004).8 My study parallels macro-theoretical contributions by Liu (2019); alongside his work, my results sug- gest that Korean targeting may not have been incoherent.My study is a first attempt to reconcile the debate in East Asia miracle with modern causal tools.
Third, I contribute to the discussion on the role of the state and development (Besley and Persson 2010, 2011; Dell, Lane, and Querubin 2018; Acemoglu et al. 2015)—especially their role in promoting industrial change (Kohli 2004). Using Vietnamese history as a case study, my work with Dell, Lane, and Querubin (2018) explores the effect of the Weberian state and its capacity to implement successful developmental policy in East versus Southeast Asia.
In sum, this study attempts to discipline a key episode of industrial policy with the toolbox of causal infer- ence. The goal is to structure coherent insights around a key, historical case of industrial transformation. By doing so, I hope to extract coherent workings of the policy—those that are useful more broadly—and to em- phasize a more empirically-grounded narrative around East Asian interventions.
I organize my study in the following way. First, section 2 discusses the institutional and historical setting of the HCI drive. Second, section 3 describes the data construction effort. Section 4 presents estimates of the di- rect impact of industrial policies on targeted industries. Section 5 turns to estimates of HCI’s spillovers into external sectors via input-output linkages. Last, section 6 concludes with a discussion of these findings.
2 Institutional Context and Political Variation
I use the historical and institutional details of South Korea’s HCI drive, which I use to identify the impact of IP in the proceeding sections (section 4). I focus on three aspects of the context: first, the geopolitical crisis that shifted in Korea’s development strategy and catalyzed HCI. Second, I detail the policy instruments that consti- tuted the intervention. Last, I describe the withdrawal of these policies, following President Park’s assassina- tion in 1979.
A) External Political Drivers A political crisis propelled South Korea’s HCI drive, which was fundamentally security driven.9 Events in the late 1960s and early 1970s created a political impasse. The first, was a sharp change in U.S. foreign policy in Asia and North Korea’s militarization (Kim 1997; Moon and Lee 2009). In 1969, President Nixon announced the end of direct U.S. military support for Asia-Pacific allies. This “Nixon Doctrine” effectively ended decades of large-scale military presence in the region. South Korea, an anti-Communist stalwart, was shocked. U.S. disengagement created the risk of full U.S. troop withdrawal from the Korean Peninsula (Nixon 1969; Kim 1970; Kwak 2003). Like their South Vietnamese allies, South Korea believed they would need to defend against a more militarized, Communist-backed neighbor. Further historical context for the nature of the “Nixon shock” is provided in the history section of the Online Appendix, including sources as to the surprise, shock-like nature of the intervention in historical literature.
8For Korea, Lee (1996) finds a negative relationship between postwar IP interventions and industry-level outcomes. Thoughtful work by Beason and Weinstein (1996) argues Japanese IP was not positively related to industry development, nor directed at sectors with scale economies. Noland (2004) similarly argues Korean policy did not target sectors with high spillovers (linkages). 9The security pretext of the policy is widely documented. “When President Richard M. Nixon declared his Guam Doc- trine in 1969 to initiate U.S. military disengagement from Asia, Park’s fear of the Americans’ departure pushed him to initi- ate an aggressive HCI drive to develop a defense industry by 1973” (Moon and Jun 2011, 119). See H.-A. Kim (2011) for HCI and the evolution of defense industry.
5 Figure 1: Political Events Surrounding HCI - US Withdrawal and Korean Provocations
Panel A) South Korean Articles on DPRK Provocations, Panel B) Mentions of U.S. Troop Withdrawal from South Korea, Top Domestic Newspapers Share of Annual Stories in New York Times 400
0.03
300
0.02 200 Number of Articles
(Major Newspapers) 0.01 100 Share of Articles Published x 100 Share
0 0.00
1961 1969 1973 1978 1961 1969 1973 1978 HCI HCI Nixon's Nixon's
Announcment Announcment
Notes: Figure shows the military−political crisis facing South Korea vis−a−vis U.S. and South Korean newspaper reporting. Panel A (left) shows the number of articles (count) in Donga and Kyunghyang newspapers matching a Korean−language dictionary of 'provocation' keywords (examples in text). I provide a list of Korean−language terms in the Online Appendix, which also details the selection of these terms using ‘‘‘word2vector‘‘‘−style models. The count includes articles matching dictionary terms that appear before page 5. The 'provocation' count in panel B matches the same count by Choi (1989) of DPRK actions violating the Korean War armistace, shown in Figure 1 of Online Appendix. Panel B (right) shows the share of New York Times news stories referring to troop withdrawal. Share is measured as the total number of full− text article hits divided by number of stories published (times 100). The search term used is 'South Korea+Troop Withdrawal', via The New York Times Lab API.
Figure 1 panel B plots coverage of U.S. troop withdrawal in the American press, measured as the share of New York Times articles containing “South Korea” and “troop withdrawal.” The first hump appears around 1970, corresponding to confirmation of American withdrawal from the peninsula. Crystallization of the withdrawal “profoundly” shocked the Park administration, who expected exemptions from Nixon’s doctrine [Trager (1972); Rogers (1970); Nixon (1970); Kwak (2003); p.34]. News attention grew through the 1971 pullout (24,000 troops, three air force battalions) from the peninsula which was seen as only the beginning of US demobilization. The second jump in panel Fig. 1 corresponds to the 1976 U.S. presidential campaign, and Jimmy Carter’s promise to end U.S. military assistance to Park. A critic of Korea’s human rights record, Carter reaffirmed his commitment after his election (Han 1978; Taylor, Smith, and Mazarr 1990; Lee 2011).10
Shifts in U.S. foreign policy came at a critical juncture for South Korea. The threat of withdrawal coincided with the militarization of North Korea and renewed provocations (Ostermann and Person 2011). Panel A of Figure 1 reports increasing antagonism by North Korea surrounding Nixon’s announcement, as reported by South Korean media. Using the full-text archive of large Korean newspapers, Dong-A Ilbo and Kyunghyang Shinmun, panel A plots the number of articles reporting on North Korean provocations. This count is constructed using a dictionary of Korean-language keywords for military antagonism (e.g. 도발(provocation), 교전(engagement), 남침(invasion of the South), 침투(infiltration), 폭파 (explosion), 포격(shelling)). The construction of this dictionary—using seed terms along with a word2vec-style model—is described in the Online Appendix. Importantly, the trends shown in panel A align with hand-collected data on military altercations by Choi and Lee (1989), shown in Figure 1 of the Online Appendix.11
10HCI’s pretext was “magnified by the Carter administration’s plan to completely withdraw U.S. ground forces.” (Kim, Shim, and Kim 1995, 186). Park’s eventual assassination complicated Carter’s commitment. 11Older versions of the paper used this data, but I turned to NLP tools since North Korea’s escalation looked quite extreme. Nevertheless, both charts supported one another.
6 Panel A conveys the wave of attacks launched by the DPRK following Nixon’s announcement, and the rising tensions before the HCI announcement in 1973. Starting in the late 1960s, South Korea experienced a string of high-profile security emergencies (Scobell and Sanford 2007).12 In the early 1970s the DPRK rivaled the South militarily. The North emerged from the Korean War with an industrial advantage, and since the 1960s pursued a total military-industrialization campaign (Hamm 1999). Relying on U.S. military support, South Korea had not done the same. During the events shown in Figure 1, South Korea had no domestic arms industry, nor the scale of industry to support it. Without U.S. troops, South Korea relied on vintage arms, and stocks incapable of absorbing a DPRK blitz (Cushman 1979; Eberstadt 1999).
These military-industrial deficiencies drove HCI policy, both in the timing and sectoral scope of policy, which I turn to now.
B) Sectoral Choice The HCI drive was announced on January 12, 1973, after a period of covert planning.13. The HCI plan is often conflated with Korea’s Third Five Year Economic Development Plan (1972-1976), which the HCI announcement effectively interrupted (Lee 1991). Using investment incentives and trade policy, HCI targeted six broad classes of strategic industry: steel, non-ferrous metals, shipbuilding, machinery, electronics, and petrochemicals (ibid; Stern et al. 1995). Table A2 lists translated names for harmonized 5-digit industries that fell under HCI priority sectors.
Two concerns dominated the choice of industry. First, HCI sectors were necessary if South Korea was to pursue military-industrial modernization, in anticipation of a future without U.S. military presence (Woo- Cumings 1998; H.-A. Kim 2011). Upstream heavy industries were the linchpin. For the Park regime, inputs like steel embodied one such critical industry (Rhyu and Lew 2011). Before 1973, the economy lacked inputs to develop a military-industrial base comparable to North Korea, which was endowed with heavy industry and pursued Communist-supported militarization. Early forays into arms manufacturing in the South were unsuccessful, with “inadequate materials and the lack of precision production. Koreans realized the critical importance of creating a more advanced industrial base” [Horikane (2005); p.375].
Second, crisis compelled Korea to choose wisely. Technocratic planning limited IP to a set of viable target sec- tors, and HCI planners used feasibility studies to winnow the scope of policies (Stern et al. 1995).14 In doing so, the regime carefully studied the industrial strategies of other small open economies (Perkins 2013). Fur- ther historical details on Korea’s technocratic planning are also provided in the History section of the Online Appendix.
Korea attempted to choose sectors for which it had a latent comparative advantage. To this end, planners saw the economy as akin to Japan—lagged. In fact, Japan’s experience was less a metaphor than a blueprint. For instance, the New Long-Range Economic Plan of Japan (1958–68) was one such technocratic blueprint, and Japan’s experience gave South Korea a template of sectors for which they had potential (Kong 2000; Moon and Jun 2011).15
C) HCI Policy Levers and Temporal Variation The HCI drive was a shift in South Korean economic policy, from generalist to targeted. Before 1973, Korea pursued a broad, export promotion development strategy
12By 1971, U.S. officials warned “our front-line is a half step before crisis” [Kim (2001); p.55]. 13See extensive treatments by Horikane (2005) and H.-A. Kim (2011) 14Jet engines and missiles, for example, were rejected as beyond their capability. For South Korea’s planning bureau- cracy, see Adelman (1969). “Planning” here refers to indicative (not central) planning inspired by France, West Germany, and Japan. 15Beyond steel and metals, shipbuilding is an example of using Japan to justify sectoral choice. “Korea found in Japan’s shipbuilding industry a cynosure” argues Woo (1991), were “the Korean strategy to promote shipbuilding was very simply a carbon copy of Japan’s” (p.137). Government documents from 1973 “dutifully note Japan’s export performance in 1955-71 and its composition of manufactures” [Kim and Leipziger (1993); p.18-19].
7 (Krueger 1979; Westphal and Kim 1982; Westphal 1990).16 The HCI-era, in contrast, was decidedly industry specific and surgical.
These surgical policies can be summarized in two broad categories: investment and production-type incen- tives, and trade policy. Investment incentives were a critical policy ingredient, notably credit for inputs and capital formation (see: Woo 1991). Starting in 1974, the National Investment Fund (NIF) promoted long-term investment with subsidized loans to HCI sectors (Koo 1984; Kim 2005).17 These “policy-oriented” loans were lent through commercial banks and state-run development banks (Koo 1984).
Figure (left) A1 shows the pattern—and volume—of lending by a key NIF lender, the Korea Development Bank (KDB), before and during HCI. The plot shows the differential pattern of lending between sectors (2-digit level). Targeted sectors are shown in red. Non-targeted are shown in black. KDB-NIF lending in Figure (left) A1 is only representative of the trend in lending however. A great deal of which was still done through the commercial banking system, which, by the HCI period, had come under various forms of state control. The line “between commercial and specialized banks [like those in Fig. A1] became blurred and both served as instruments” for government directed credit [Cho (1989); p.93].
The right panel of Figure A1 provides another view of incentives over the HCI period, and conveys differ- ences in the cost of capital across sectors. Specifically, the right panel plots estimates of effective tax rates on the returns to capital, accounting for the package of industry-specific subsidies through the HCI drive (calculated by Kwack 1985; Stern et al. 1995). Thick lines show average rates in HCI versus non-HCI sectors. Thin lines are average rates disaggregated at the 2-digit industry level. Post-1973, the rates across industries diverge markedly, reflecting the HCI drive’s sectoral bias (Kwack 1984; Kim 1990).18
Second, trade policy shifted from general export promotion to HCI promotion. Pre-1973, exporters enjoyed a “virtual free trade regime” and were exempted from a number of import restrictions on inputs (Nam 1980; Westphal 1990). HCI ended this early program, eliminating the allowances and import subsidies granted, broadly, to exporters. Post-1973, however, HCI industries enjoyed 1960s-style import assistance and exemptions (Woo 1991; Cho and Kim 1995). For example, HCI producers were exempted from up to 100 percent of duties and tariffs on imports.19
HCI had an unexpected expiration date. On October 26, 1979, President Park was assassinated by Korean Cen- tral Intelligence Agency director, Kim Jae-kyu. The assassination ended Park’s regime and its keystone indus- trial drive (Cho and Kim 1995; Lee 1980). The successor regime rejected the dirigisme of the Park era and ratio- nalized HCI policy in repeated rounds of economic liberalization.20
Thus, post-1979, South Korea dismantled HCI incentives, pursuing “investment adjustment” for targeted sectors and further trade liberalization (Kim 1988, 1994; Kim and Leipziger 1993). The import liberalization ratio climbed from 68.6 in 1979 to 76.6 by 1982, and restrictions were further reductions between 1982 and
16Conditional on manufacturing, this policy was broad. As these export incentives “were administered uniformly across all industries” [Westphal and Kim (1982); p.217-218]. For example, the main role of credit policies “was to support export ‘activity’ rather than specific industries” [Cho (1989); p.93]. 17The NIF was largely funded through bond sales of financial institutions. According to Byung-kook Kim, “NIF was an outright forced savings program,” selling bonds on public non-banking institutions and then requiring 8 percent of wage income to be levied into pensions [B.-k. Kim and Vogel (2011); p.226]. 18Previous export tax incentives “no longer played a central role compared to that played by [the] industry incentive scheme,” which directed investment to “a relatively small number of industries” [Trela and Whalley (1990); p.19]. “Spe- cial Tax Treatment for Key Industries” under the Tax Exemption and Reduction Control Law was one such example of HCI investment incentives. 19According to Park (1977)’s calculation for the period “key industries,” on average, enjoyed 80 percent tariff exemptions across industries (with the exception of petrochemicals) (p.212). Meanwhile, HCI exporters were allowed to purchase inputs from foreign investors and licensors (Castley (1997)). 20HCI was not without its cracks. Earlier in 1979, the government had announced the “Comprehensive Stabilization Pro- gram,” in efforts to address the apparent macroeconomic instability brought on by turbulent world economic conditions and exacerbated by HCI lending. Nevertheless, Park’s removal meant wide scale retrenchment of HCI in earnest.
8 1984. The banking sector was also liberalized, with notable reforms in 1981 and again in 1983. The share of government “policy loans” to industry shrank, and interest rates between strategic and non-strategic sectors converged (Cho and Cole 1986; Nam 1992). By 1982, the gap in effective corporate tax rates between strategic and non-strategic industries was also closed (Kwack and Lee 1992) (reported in A1).
3 Data Construction
I construct new data on industrial development over South Korea’s miracle period (1967-1986). Below I describe the construction and harmonization of this data. Since, however, harmonization required nearly a dozen crosswalk schemas, I provide additional technical details in the data section of the Online Appendix.
Table A1 shows pre-1973 statistics (mean and standard deviation) for core variables. For visibility, A1 mostly reports non-normalized values. Throughout this paper, however, I default to inverse hyperbolic sine (ihs throughout) normalization, which accommodates 0s and negative values. Variables, such as disaggregated investment data, contain zeros, and inventories contain both zeros and negative values. This is especially prescient in higher resolution industrial panels.
The Long and Short of Industrial Panels The empirical strategy of this paper requires following industries consistently through time. I construct industry-level panels using digitized data from the Economic Planning Board’s (EPB) Mining and Manufacturing Surveys and Census (MMS). True to its name, the HCI drive was a fundamentally industry-level intervention. To this end, MMS industry data is apt for studying sectoral policy. It is high quality; reports manufacturing outcomes across the study period; and importantly, it is available at a fine level of disaggregation.
The MMS published census data nearly every five years (1970, onward, with annual intercensal surveys) at the 5-digit industry level, and aggregated from establishment-level enumeration.21 I supplement digitized MMS statistics with vintage MMS tape data (1977–1986) sold by the EPB in the 1980s. Additional data on prices are digitized from historical and contemporary Bank of Korea producer price index publications.
Creating consistent industry panels is non-trivial. Specifically, following industries through the HCI period (and beyond) requires extensive concordance and harmonization. Between 1967-1986, the horizon of this study, the EPB updated Korea’s industrial codes (KSIC, based on the international ISIC standard) four times, with a major revision in 1970.
Thus, I use two harmonized panels throughout this study, which are shown in Table A1. Part A of Table A1 reports values from the shorter, granular 5-digit industry panel, which is harmonized from 1970 to 1986. Part B reports values from the longer, but much more aggregated, 4-digit panel. This data is harmonized from 1967 to 1986. I choose 1986 as the end date of the panel, as it is when my mainframe data series ends and it is the year before the disrupting transition to democracy in 1987.
The harmonization process accounts for industry code changes in official Korea manufacturing data and in- troduces a trade-off.22 On the one hand, shorter panel data (1970-1986) accommodates more industry observa- tions, but does so over a smaller period. Thus, 5-digit data requires less harmonization as it is closer to origi- nal MMS values. On the other hand, longer data (1967-1986) covers fewer industries, but does so over a longer period. Thus, 4-digit data requires more harmonization (thus aggregation), but encompasses critical pre-HCI periods.
21To illustrate this level of aggregation consider two sectors: 35291, Manufactures of adhesives and gelatin products, and 35292, Manufactures of explosives and pyrotechnic products. Note that because the census is enumerated at the establishment- level, as opposed to the firm, this precludes analysis of firm competition. Micro data is not readily available for this early period. 22Fundamentally, this harmonization process entailed digitizing and rebuilding official crosswalks. I provide details of the process in Data Appendix.
9 Linkage and Trade Data Data on sectoral linkages come from the Bank of Korea’s 1970 “basic” input-output tables. These are the most disaggregated available for the period, and cover around 320 sectors. Hard copies were translated to English and then digitized.23 I also construct Leontief input-output coefficients from 1970 tables, which I describe in section 5.1.
Bank of Korea data and MMS industrial surveys use different coding schemes. This means that combining input-output accounts with industry data requires further harmonization. The Online Data Appendix describes this process, along with other harmonization and cross-walk procedures required by my analysis.
I supplement the 1970 input-output tables with 1975 tables. Though produced during the HCI drive, the 1975 tables distinguish between domestic and imported intra-industry flows. I return to this distinction in the net- work analysis section 5.1. Throughout the study, I default to referring to the “general” 1970 linkages, unless otherwise specified.
I use international trade flow and trade policy data as a standalone panel. This trade panel is also mapped to the “long”, 4-digit industrial panels described above. Trade flows come principally from the UN-COMTRADE database and are originally reported at the 4-digit SITC (Rev.1) level. Thus, I also construct crosswalks to con- nect trade-level data to Korean industrial data.
Trade policy—product-level measures of quantitative restriction (QRs) coverage and tariffs—are digitized from Luedde-Neurath (1986) (at the Customs Commodity Code Number [CCCN] level). This data is available for 1968, 1974, 1976, 1978, 1980, and 1982. This is the most disaggregated, readily-available data for the period (Westphal 1990). Moreover, these statistics are notable in that they contain measures of core non-tariff barriers, notably quantitative restrictions (QRs) (Goldberg and Pavcnik 2016).24
I use trade policy data to calculate separate measures of output and input market protection.25 Output pro- tection for industry 8 is simply the average tariff or QR score for that sector: output-tariff8. HCI used exemp- tions from import barriers as a policy tool. Thus I calculate measures of input protection. Input tariffs faced by 8 (or QRs) are calculated as the weighted sum of tariff (QR) exposure, with weights taken from the 1970 I- O accounts (following Amiti and Konings 2007). As such, exposure is calculated as input-tariff = Í 8 9 98 × output-tariff9, where 98 are input cost-shares for industry 8.
I now consider how the HCI industrial policy bundle might propagate through the input-output network.
4 Direct Impacts of Industrial Policy
4.1 Empirical Framework
This section explores how HCI industrial policy impacted the evolution of targeted industry—during and after the policy period. I use this variation to estimate the impact of HCI industrial policy. This section describes how. First, I describe how the HCI context is useful in estimating the impact of industrial policy. Second, I show how this variation informs a differences-in-differences estimation strategy.
Empirical Import of HCI Variation The HCI episode sharply altered the sectoral bias of Korean industrial strategy, while its introduction and withdrawal were driven by external political forces. From an empirical point of view, the context of the HCI is a useful experiment for estimating the impact of industrial policy.
23At the time of this study, machine readable I-O tables for 1970, were not available from the Bank of Korea. Note that 1970 tables report total values of inter-industry flows and do not differentiate between domestic and imported activity. 24Archival administrative data has been collected as of this study. Most empirical studies of Korean trade policy use highly aggregated data. I describe Luedde-Neurath (1986)’s coding of QRs in the Data Appendix. 25For simplicity, this study follows the contemporary practice of using nominal, statutory measures, though I also calculate measures weighted by input share for the corresponding year.
10 First, in practice industrial policy is often negatively correlated with industry fundamentals. Clientelism and capture mean policies go to industries that contradict comparative advantage (Rodrik 2005; Lin 2012). Yet be- yond rent-seeking, it is common for IP to support stagnating industries (“sunset” IP as opposed to infant in- dustry IP).26. While sunset policies or industry support are often driven by political demand, they may also be socially optimal (Hillman 1982; Flam, Persson, and Svensson 1983). Given this policy orientation, empirical work often shows a negative relationship between IP and industrial development outcomes (Harrison 1994; Harrison and Rodriguez-Clare 2009; Rodriguez and Rodrik 2001).
HCI, however, is an infant industry policy aimed at sectors for which South Korea has unrealized—or latent— comparative advantage. Critically, the HCI episode distinguishes between sunrise and sunset policies.27 Sim- ilarly, HCI was a top-down shift in national industrial strategy, whose sectoral bias was not driven by politi- cal constituencies. President Park’s consolidation of power allowed the creation of a technocratic Heavy and Chemical Industry Planning Board that transcended competing state factions.
Second, the external politics of HCI meant that de jure policy was binding. In practice, industrial policy pro- nouncements often depart from actual implementation. Even where policy is coherent and deployed, polit- ical constraints can undermine the quality of administration. Subsidized credit may not be directed toward prioritized industry (Lazzarini et al. 2015; Musacchio, Lazzarini, and Aguilera 2015), and trade policies may be undermined by capacity constraints (Panchamukhi 1978). Thus, regressing outcomes of statutory IP mea- sures may not reveal the impact of policy, but the quality of implementation. The politics behind HCI, how- ever, meant policy was incentive compatible. Expedient political factors made HCI binding—both its imple- mentation and removal.
Third, the HCI drive (1973-1979) was temporary. The theoretical case for infant industry policy often involves temporary policy (Melitz 2005). Historically, however, temporary policies often become permanent (Juhasz 2018), complicating empirical tests. For HCI, the 1979 assassination of President Park effectively ended HCI. I use the post-1979 period to examine the long-run impacts of the policy.
Fourth, HCI was a purposeful shift in national industrial policy toward a discrete set of infant industries. Given the complications of estimating the impact of IP, researchers have used valuable natural experiments that mimic policy variation (e.g. Juhasz (2018) and Hanlon (2020)). Broadly, however, the case for IP hinges on policy being allocated intentionally—according to technocratic criteria. This rationale can make it difficult to glean insights from random variation that does not embody this criteria (Rodrik 2004).28 The HCI episode provides useful temporal and sectoral variation for studying industrial policy “in the wild.”
The HCI episode thus provides a useful setting to estimate the short and longer run impacts of industrial pol- icy. HCI industries selected were nascent industries, as opposed to sunset industries. The geopolitical crisis meant IP was purposeful; planned in secret; and implemented, top-down, by a government, limiting the scope of domestic lobbying. A planning apparatus reduced—intentionally so—the potential of choosing sectors that contradicted notions of comparative advantage.
Estimation The temporal and sectoral variation of HCI informs a natural differences-in-differences style strat- egy. I use the January 1973 announcement of HCI as the start date, and 1979 as the de facto end date of HCI. The end date corresponds to President Park’s assassination, which promulgated the repeal of IP incentives by the successor regime.
26Such policies may be permissible under multilateral agreements. 27For example, while Korea’s HCI supported a “sunrise” shipbuilding industry, their Swedish contemporaries supported a “sunset” shipbuilding industry. 28Criscuolo et al. (2019) and Giorcelli (2019) illustrate the possibilities of using random variation in ways that, neverthe- less, address these issues.
11 Using a dynamic differences-in-differences design, I estimate the differences between targeted HCI industries and non-targeted manufacturing industries, each year relative to 1972. I examine these differences, before, dur- ing, and after the policy. Formally, I estimate the following
Õ 9 Õ 9 . = C Targeted Year -0 Year Ω & 8C 8 + + 9 · 8 × C + 8 × C 9 + 8C (1) 9≠1972 9≠1972
8 indexes each 5-digit or 4-digit manufacturing industry, and C denotes the year. For the long 4-digit panel, C takes the values 1967 1986, and 1970 1986 for the more granulate 5-digit panel. Outcome . is an industrial − − development outcome. For industrial data (see section 3) I will make ample use of the inverse hyperbolic sine transformation (ihs) for outcomes.
The main effects of the policy are estimated using the term, Targeted, a binary variable equal to if HCI targets an industry, and zero otherwise. This “sharp” treatment term has a number of benefits. First, this simple indi- cator allows me to visually assess counterfactual industry dynamics, by clearly plotting pre-trends and group averages (see results section Figure 2). Second, it allows clearer comparisons with the semi-parametric estima- tors described below, which require binary treatment. Third, binary treatment easily codifies HCI targeting for industries that fall under the drive’s list of key strategic industries.
The baseline equation (1) is a linear two-way fixed effects (TWFE) specification. Industry fixed effects,
8, mean that I use variation within each industry. Time effects, C, control for common temporal shocks. Standard errors are corrected for heteroskedasticity and clustered at the industry-level, allowing for within-industry correlation.
I include a set of baseline pre-treatment controls which attempt to capture unobserved productivity correlated with the intervention. These include measures of total material costs, average wage bill, average plant size, and labor productivity. Since these averages do not change through time, I interact them with period effects, - Year. Thus, their effect is allowed to flexibly vary through time. 80 ×
The coefficients of interest, 9, describe the evolution of manufacturing industries through time. The set of these 9s are the differences between targeted and non-targeted industries for each year 9, relative to the pre-treatment year 1972, with the coefficient for 1972 normalized to 0.
The set of 9 convey three aspects of how HCI industries evolved between 1967 and 1986.
First, estimates after 1972 show the average impact of the policy package for each period after the start of HCI. If HCI is associated with short-run (that is, during the six year drive) industrial development, we should ob- serve increasing differences in . between 1973 and 1979: 9 1973, ... . ≥ ˆ1974 ≤ ˆ1975 ≤ ≤ ˆ1979 Second, estimates after 1979 describe the more durable, long-run impacts of the temporary infant industry policy. In the parlance of the IP literature, the longevity of these effects indicate the potential “dynamic effects” of IP. Be they through dynamic scales economies, forms of learning-by-doing, or other forces that transform temporary advantages into persistent effects. For 9 > 1979, estimates ... , would be ˆ1979 ≤ ˆ1980 ≤ ˆ1986 compatible with a more persistent, or dynamic impact of the policy.
Third, estimates before 1972 describe average differences between HCI and non-HCI industry before the policy. Thus, they convey information—with caveats—for the common trend assumption of the research design. Pre-1972, we should not observe systematic differences between treatment and control industries:
12 0. Thus, I provide joint tests for whether are not equal to 0 over the pre-treatment ˆ1967 ≈ ˆ1968 ≈ ˆ1972 ≈ 9 period.29
Accordingly, the identifying assumption behind equation (1) is that differences in industrial development be- tween treated and non-treated industries would have evolved similarly in the absence of the policy. I assume the comparative advantage of HCI industries was latent, and thus had not materialized—and may not have, absent the HCI policy bundle.
Although pre-treatment movements in targeted sectors is helpful to this end, the analysis in the second part of this paper may also be useful (section 5.3.2). Movements—or lack thereof—in industries downstream or upstream from targeted sectors can potentially reveal aspects of the pre-1973 HCI environment that may not be directly observable in HCI sectors themselves. Movements in external sectors—upstream or downstream— may be informative as to supply-side or demand-side trends that may confound estimates of HCI’s impact over the period.
Thus far, I have focused on the impacts of HCI for each point in time. Yet, it is useful to consider the average impact of HCI an either side of the intervention—before and after 1973. Consider a simplified specification,